THE JOURNAL OF INDUSTRIAL ECONOMICS
Volume XLV
June 1997
0022-1821
No. 2
PRICE WARS AND THE STABILITY OF COLLUSION:
A STUDY OF THE PRE-WORLD WAR I BROMINE
INDUSTRY*
Margaret C. Levenstein
Between 1885 and 1914 US bromine producers colluded to raise prices
and pro¢ts. This collusion was disrupted by price wars. Bromine price
wars are compared with the Green/Porter and Abreu/Pearce/
Stacchetti models. Some price wars resulted from the imperfect
monitoring problems which motivate these models. Several empirical
implications of the APS model are borne out, but the bromine
industry's price wars were generally milder than contemplated by APS.
More severe price wars were part of a bargaining process, in which
¢rms tried to force renegotiation to a new collusive equilibrium with a
di¡erent distribution of rents.
i.
introduction
Economists have long debated the causes and consequences of price wars.
Stigler [1964] argued that cartel pro¢ts induce cheating or entry. A price
war inevitably follows, and collusion disintegrates. Earlier students of
industrial organization believed, to the contrary, that the possibility of
price wars stabilized collusion (Burns [1936], pp. 27^30). When price wars
arose, they were ascribed to exogenous forces disrupting otherwise stable
cartels (Stocking and Watson [1951], p. 505). Hearkening back to the older
literature, game theorists have argued that price wars may be necessary
for cartel stability, and that they occur in response to exogenous shocks
(Green and Porter [1984], Abreu, Pearce, and Stacchetti [1986], hereafter
APS). Firms engage in price wars to deter cheating, which, in a world of
imperfect information, they cannot distinguish from stochastic
£uctuations. These ``equilibrium price wars'' do not re£ect the collapse of
collusion, but are instead a crucial element in its stability.
These game-theoretic models have not been widely tested, partly
because they depend on variables, like beliefs and strategies, that are
* The author acknowledges helpful comments from two anonymous referees, David Arsen,
Severin Borenstein, Je¡ Borland, Insong Gil, Thomas Iosso, Francine Lafontaine, Kevin
Lang, Richard Levin, Nancy Lutz, Paul Milgrom, David Pearce, Matthew Shapiro, David
Weiman, Lawrence White, and participants in seminars at the University of Chicago,
Stanford University, UCLA, and the Cliometrics Society Summer Conference. She also
appreciates the valuable assistance she has received from E. N. Brandt, Barbara Brennan,
Delores Goulet, Margaret Lyon, and Kathleen Thomas, all of the Post Street Archives, and
from the archivists at the Baker and New England Depository Libraries. All errors, of course,
remain her own.
ß Blackwell Publishers Ltd. 1997, 108 Cowley Road, Oxford OX4 1JF, UK, and 350 Main Street, Malden, MA 02148, USA.
117
118
margaret c. levenstein
not observable. In this paper I analyze internal documents from a turn
of the century cartelöcontracts, correspondence, minutes of
negotiations found in the archives of the Dow Chemical Companyö
that explicitly discuss what industry participants did and did not
believe, which strategies they pursued and which they rejected. I use
these sources, as well as industry and government publications, to
evaluate the role of price wars in stabilizing, and destabilizing, the
bromine cartel.1
The bromine cartel in the US did use the threat of price wars to provide
individual ¢rms with an incentive to participate in the cartel, as suggested
by Friedman [1971]. An examination of the historical incidents leading
up to each of the six bromine price wars suggests that imperfect
monitoring problemsöthe inability to distinguish between random shocks
and real cheatingödid contribute to the occurrence of some price wars.
A comparison of the qualitative characteristics of price wars with the
optimal punishment described by APS ¢nds that in four cases the initial
price cut and the duration of the price war were similar, as predicted.
However, those price wars were relatively mild, and did not approach the
Cournot reversion levels proposed in Green and Porter and Friedman, let
alone the more severe punishments contemplated in APS. Two other price
wars began right after publicly announced violations of the collusive
agreement. These price wars were the result of disagreements among
participant ¢rms regarding the distribution of rents generated by collusion.
These price wars lasted longer, and prices fell much further. These two
price wars are better understood as breakdowns in collusion than as
equilibrium punishments.
ii.
the bromine pools
Following the US Civil War battle¢eld discovery that potassium
bromide had sedative properties, it quickly became a popular patent
medicine, used as a headache remedy and stomach settler. While
elemental liquid bromine had been produced commercially in the
United States since 1845, demand had been limited to use in
daguerreotype photography. In response to the new pharmaceutical
demand, sodium chloride salt producers whose underground brine
contained bromine began extracting it to make elemental liquid
bromine. These ¢rms pumped and heated the brine, extracting sodium
chloride and then bromine. Virtually all of this elemental liquid
bromine was then sold to one of several large pharmaceutical
1
Detailed citations from these documents can be found in Levenstein [1993].
ß Blackwell Publishers Ltd. 1997.
price wars and collusion in bromine pre-world war i
119
distributors who converted the bromine into potassium bromide salts.
These pharmaceutical distributors had national sales networks through
which they distributed potassium bromide salts to patent medicine
producers, pharmaceutical ``jobbers,'' and pharmacies around the
country. In 1880 there were twelve liquid bromine producers, located
along the Ohio and Kanawha Rivers. (These ¢rms will be referred to,
collectively, as the ``Ohio River bromine producers.'')
From 1885 to 1902, competition in the upstream bromine industry
was controlled by a pool. The pool was an independent, unincorporated ¢rm with contracts to buy the entire output of every bromine
producer. These contracts guaranteed producers a price, and prohibited
them from selling to anyone else. The contracts were explicitly
conditioned on the participation of all producers. From 1885 to 1891
the pool operated as the National Bromine Company (NBC). From
1892 to 1902 it was run by W. R. Shields. The pool itself had
contracts to sell all its bromine to two of the largest pharmaceutical
distributors in the country, Mallinckrodt Chemical and Powers &
Weightman (P&W). The pool also had an agreement with its only
potential foreign competitionöthe Deutsche Bromkonventionöreserving
the United States as the exclusive territory of the American producers,
in return for its refraining from exporting.
The structure of collusion changed somewhat in 1894, when the
downstream ¢rms, Mallinckrodt and P&W, together contracted directly
with a new entrant to purchase all of its output. This ¢rm, the Midland
Chemical Company (predecessor of the Dow Chemical Company),
introduced a patented technology that integrated the production of liquid
bromine and potassium bromide salts, and had signi¢cantly larger
minimum optimal scale. Midland's contract restricted its output to
100,000 pounds of potassium bromide a year, about twice the equivalent
output in bromine of a typical Ohio River ¢rm. Thus, Midland vertically
integrated the upstream and downstream production processes, but did
not integrate into the distribution function performed by the downstream
¢rms. Shield's contracts with the Ohio River bromine producers and with
Mallinckrodt and P&W remained in force.
The structure of collusion changed more dramatically in 1902 when
Dow, having purchased Midland in 1900, refused to renew the
contracts Midland had signed with Mallinckrodt and P&W. It
contracted instead with two other distributors whom Dow selected
because they had a reputation for not cutting prices. Dow hoped that
this selection of outside distributors would prevent a decline in prices
despite its decision to operate independently of the pool distributors.
Shield's contracts with the Ohio River ¢rms expired and were not
renewed, though he continued to act as an intermediary between the
Ohio River and Mallinckrodt and P&W. After a brief price war, an
ß Blackwell Publishers Ltd. 1997.
120
margaret c. levenstein
Figure 1
informal collusive arrangement was arrived at, in which all ¢rms
promised not to undercut the announced potassium bromide price in
the leading industry publication, the Oil, Paint and Drug Reporter. The
market division agreement with the Germans was also renewed, but
Dow soon began secretly exporting bromides to Europe. In 1905 the
Germans retaliated by entering the US market at ¢fty percent of the
going price.
Cooperation was re-established at the end of 1908. Dow and the
Germans agreed to set prices for all bromine products jointly and divide
world markets. Cooperation within the American market was also reestablished. Dow signed distribution contracts with all the major
potassium bromide distributors. These contracts prohibited the
distributors from buying bromine from anyone but Dow. For over a year,
Dow kept the price of potassium bromide low enough that the Ohio River
¢rms, with their higher production costs, could not pro¢tably sell their
bromine. In 1910 Dow contracted with all but one of the Ohio River ¢rms
to purchase their entire output on a sliding scale price, based on the
published price. Thus Dow recreated the pool structure, but itself acted as
the pool, purchasing the entire output of its competitors. The price of
potassium bromide rose slowly but steadily until the beginning of World
War I (Figure 1).
ß Blackwell Publishers Ltd. 1997.
price wars and collusion in bromine pre-world war i
iii.
121
evaluating the success of the cartel
Before analyzing in more detail the function of price wars in the bromine
industry, I address a more basic question. Was collusion successful? Did it
restrict output, raise prices, and increase the pro¢ts of participating ¢rms?
And if so, by how much? An analysis of price trends and a comparison
of actual prices with those achievable without cooperation supports the
contention of this paper, and the belief of the participants in the industry,
that cooperation increased prices and pro¢ts.2
III(i).
Price Trends
The price of bromine fell continuously after the commencement of commercial production, from six dollars a pound before the Civil War to thirty
cents in 1875. It fell 7% between 1875 and 1880, and almost 30% between
1880 and 1884.3 With the establishment of the bromine pool in 1885, this
trend reversed. The price of potassium bromide increased 23% over the year
(Figure 1).4 The average price during the NBC pool (1885^1891) was almost
ten percent higher than the average price of the previous ¢ve year period.
When the NBC contracts terminated in 1891, prices returned to their prepool pattern, falling by almost thirty percent (Figure 1). With the
establishment of the ``Shields pool'' in October 1892, prices began a steady
upward climb. The average price during the Shields pool was 60% higher
than during the period between the pools.5 At the dissolution of the Shields
pool in 1902, the price again fell quite sharply (43.2% in one month).
2
Because their account books have not survived, direct pro¢t measures for the Ohio River
¢rms and P&W and Mallinckrodt are not available. Evidence of the Midland Chemical
Company's pro¢tability is available but it is di¤cult to separate pro¢ts accruing to cartel
prices from those accruing to Dow's patented inventions. Between 1894 and 1900, potassium
bromide sold to Mallinckrodt and P&W was Midland's sole product. Midland issued its ¢rst
dividend after signing its contract with P&W and Mallinckrodt in 1894. Midland paid
monthly dividends of two to ¢ve percent on the par value of its stock over the next six years
(Levenstein [1997]). Annual interest rates in the region averaged less than 6.5% (Davis
[1965]).
3
The price of elemental liquid bromine fell from between six and eight dollars per pound
in 1845 to $4.50 per pound in 1868, to 70¿ per pound in 1870 (Haynes [1954], p. 324 and
OPDR [1880], p. 655].
4
After the NBC contracts went into e¡ect, virtually all elemental liquid bromine traded at
contract prices. The bromine price quotations in the OPDR were for small lots of bromine
purchased from the NBC, not the Ohio River producers. For market prices, I rely on
quotations of potassium bromide prices. The average price of potassium bromide between
February 18, 1880 and February 11, 1885 was 31.07¿. The comparable price during the NBC,
February 18, 1885 to March 4, 1891, was 34.08¿.
5
The average price of potassium bromide during the Shields pool (October 13, 1892 to
October 20, 1902) was 40.88¿, 60% above the average price of 24.75¿ from March 11, 1891 to
October 3, 1892. Since the low price in between the two pools could re£ect pricing strategies
below competitive prices, the relevant comparison may be to the pre-pool prices of 18801885. The increase associated with the Shields pool would in that case be 31.6%.
ß Blackwell Publishers Ltd. 1997.
122
margaret c. levenstein
After 1902, Dow's large share of total output, produced at lower cost,
makes comparisons with pre-1885 prices less meaningful. The average
price after 1902 was lower, during both cooperative and non-cooperative
periods, than during the preceding decades. The di¡erence between prices
during cooperation and non-cooperation remains. Between 1902 and 1914,
the average price during cooperation was almost 75% higher than during
non-cooperative periods.6 Similarly, the price increased by almost 50%,
from 13¿ to 19¿, on December 21, 1908, when the Dow-Deutsche
Bromkonvention agreement took e¡ect. The price trend for the previous
three years had been £at or negative (Figure 1). The price remained at 19¿
until May 1910, when the extant Ohio River ¢rms agreed to sell their
output to Dow. Immediately thereafter prices began an upward trend that
continued until World War I. By 1914, nominal prices had returned,
despite decreases in costs, to the levels charged prior to the breakup of the
pool in 1902 (Figure 1).7
The stability of prices during cooperation was greater than during noncooperative periods. This does not itself indicate that prices were higher,
but is consistent with their being set non-competitively (Slade [1991]). The
market power exerted by cooperating ¢rms allowed them to ignore small
£uctuations in demand or cost. Cooperation may have required infrequent
price changes, which had to be coordinated among several, geographically
dispersed ¢rms. During non-cooperative periods, ¢rms independently
announced prices, increasing the probability that a price change would
occur in any given week. Prices changed every 11.5 weeks during noncooperative periods but only every 41.1 weeks during cooperation. The
results are similar if the pool (1885^1902) and post-pool (1902^1914) data
are separated. During the pools, prices changed every 35 weeks during
cooperation, every 6.8 during non-cooperation. After 1902 the pattern
remained; prices endured 64.7 weeks during cooperation, 17.5 weeks
during non-cooperation.
Average prices were higher during periods of cooperation. Prices
increased during cooperation, and fell at its conclusion. Price £uctuations
were greater during non-cooperative periods, re£ecting the independent
price setting behavior of ¢rms. Each and every change in price trends can
be explained by a change in the cooperative behavior of the ¢rms in the
industry. There is no evidence that changes in demand or cost can explain
the observed £uctuations in price.
6
The average price of potassium bromide during non-cooperative periods was 16.8¿.
During cooperation it was 29.3¿.
7
All of these results hold if one examines real instead of nominal prices. See the Journal's
editorial Web site.
ß Blackwell Publishers Ltd. 1997.
price wars and collusion in bromine pre-world war i
III(ii).
123
A potassium bromide cost function
To evaluate more fully the e¡ectiveness of the collusive policies pursued
by bromine producers and distributors, I next compare the prices charged
during the period studied to counterfactual prices which would have been
charged in a competitive regime. This requires estimates of the demand
and cost functions for the industry. Unfortunately, I have cost data for
neither the Ohio River producers nor the pool's distributors. But Dow's
records do provide information on the parameters of its competitors' cost
functions. Where there is ambiguity, I use the highest cost parameters so
that, if anything, the e¡ect of the cartel is underestimated.
Consistent with discussions among participants in the industry, I
assume that upstream ¢rms manufactured liquid bromine with constant
marginal cost …MCBr † and, implicitly, that the bromine producers had a
capacity constraint which was determined by plant size and the ¢rm's
output of sodium chloride salt. Direct measures of MCBr are not
available, but Dow records contain several estimates. These estimates
were made at three di¡erent times, in considering plans to gain control
of Ohio River bromine. In 1896 and 1897 the Midland Chemical
Company considered ``dead renting'' the Ohio River producers, paying
them not to produce bromine. Midland attempted to determine the
pro¢t they were making on their bromine, so that Midland could o¡er
them an equivalent amount not to produce. Midland knew the price that
they received from Shields; the only dispute was the average cost of a
pound of bromine. Shields, representing the bromine manufacturers in
these negotiations, tried to convince Midland that MCBr was low.
Herbert Dow and Henry Cooper, Midland's general manager, opposed
the policy of ``dead renting'' and argued that MCBr was much higher.
They argued that MCBr could be as high as 18¿ to 21¿, saying that when
prices had fallen much below that the Ohio River ¢rms had stopped
producing. J. H. Osborn and B. E. Helman, Midland's vice president
and treasurer, respectively, advocated dead renting and often quoted
Shield's estimates of 10¿ to 12¿ to Dow and Cooper. By 1902, Herbert
Dow seems to have accepted that MCBr was about 12¿ per pound. That
is the ¢gure he presents to the ¢rm's directors when discussing the
implications of Dow's decision to end its cooperation with the pool. He
gives the same estimate to outsiders. The subject is raised again in
1910, when Dow contracted to buy the output of the Ohio River ¢rms.
The contract had a sliding scale price for liquid bromine between 11¿
and 19¿ per pound (depending on the selling price of potassium
bromide). As the Ohio River ¢rms were not required to produce or sell
any amount under these contracts, it is still possible that MCBr was
above 11¿. So as not to underestimate their costs (and overestimate the
e¡ects of the cartel), the calculations presented below use what came to
ß Blackwell Publishers Ltd. 1997.
124
margaret c. levenstein
be the consensus opinion at the Dow Chemical Company, 12¿ a
pound.8
I assume that the downstream ¢rms converted bromine into potassium
bromide with a Leontief, ¢xed proportions technology, so the cost
function for converting bromine to potassium bromide has the form
MCKBr ˆ bPbromine ‡ CPpotash ‡ t
where b and C are technical coe¤cients measuring the liquid bromine
and potash contained in a pound of potassium bromide and t is the cost
incurred converting bromine into a pound of potassium bromide. Pbromine
and Ppotash are the prices of potash and bromine. The latter was published
weekly in the Oil, Paint and Drug Reporter, and its average cost was four
cents per pound. While a bromine price was published in the Oil, Paint and
Drug Reporter, the relevant price is that paid by Mallinckrodt and P&W,
and speci¢ed in pool contracts. I do not have the NBC contract prices, but
I do know those in Shields' contracts, and they are used below.9 The
parameters of the potassium bromide cost function …b; C and t) are
determined from estimates in the Dow correspondence, as well as Dow's
own production records and an estimate made, in 1910, of what it would
cost Dow to convert Ohio River bromine to potassium bromide.
In the 1896^97 correspondence, Osborn and Helman give, at di¡erent
times, ¢gures implying a b of .7 and .75. In 1910, Shields told Dow that
there were 85 pounds of bromine in a hundred pounds of potassium
bromide. Herbert Dow's 1910 estimate of the cost to Dow of converting
Ohio River bromine assumed a b ˆ :8. A few months later Dow wrote to
the ¢rm's German distributor that there were approximately two thirds of
a pound of bromine in a pound of potassium bromide. Since the highest
8
The Ohio River manufacturers all produced bromine as a by-product to the manufacture
of salt, so changes in the price of salt might a¡ect the output of bromine. The price of salt was
falling over the entire period covered by this study, and salt producers in the Ohio River region
were facing increasing competition from new sources of supply (Levenstein [1995]). The
primary impact of this decline in salt prices was the exit of several ¢rms over the period. While
the calculations below assume that there were twelve producers, by 1910 there were only eight
active domestic producers. The combination of low salt prices and low bromine prices during
the 1905^1908 price war and Ohio River £ooding caused the exit of four producers. Because
the production technology required that salt be extracted before bromine, I, like Dow, treat the
cost of salt production as a ¢xed cost. Herbert Dow described the cost of bromine as 12¿ per
pound, ``plus the loss on salt'' (memorandum, 27 January 1902). When the Midland was
thinking of ``dead renting'' the Ohio River bromine plants in 1896 it had discussions with the
United Salt Company, an Ohio salt pool, with the hope that it would contribute to the cost of
shutting down competitor plants. Acknowledging the ease of entry in salt production, United
Salt decided not to pursue the dead renting strategy. The price of salt, and its pro¢t or loss to
the Ohio River ¢rms, is never again discussed in Dow documents, suggesting that it had little
e¡ect on the marginal output decisions of the Ohio River ¢rms.
9
Shields purchased the entire stock on hand in 1892 at 14¿ per pound. The Ohio River
producers received 16¿ per pound during the ¢rst year of the contract, and a 2¿ increase each
year (up to 22¿).
ß Blackwell Publishers Ltd. 1997.
price wars and collusion in bromine pre-world war i
125
estimate of b (.85) comes from Shields, who generally had an incentive to
understate the costs of making potassium bromide from Ohio River
bromine, and since I am attempting to bias the results away from large
cartel e¡ects, the cost function below uses that estimate.
The estimates of C, the amount of potash in a pound of potassium
bromide, in the Dow correspondence vary. Fortunately, Dow's own
production records themselves augment these. While Dow's production
process di¡ered from that used by its competitors, the physical
requirements of potash were approximately the same. As was the case
for the other parameters, the lowest estimate of C, .25, comes from
Osborn. In concurrent correspondence, Helman reports numbers from
Shields which imply a C of .44. These ¢gures are substantially below
that implied by Dow's own production records. Production records
from Dow's two bromides plants for both 1905 and 1906 each give a
C of approximately .58. In Dow's 1910 estimate, he assumed that .57
pounds of potash were necessary to make a pound of potassium
bromide from bromine. The conversion cost function rounds these
highest estimates up to give a C of .6.
There is even less information available about the parameter t, the cost
incurred by Mallinckrodt and P&W in converting bromine to potassium
bromide. Even Shields probably had no direct information about these
costs, and Dow's correspondence with Mallinckrodt and P&W never
mentions their conversion costs. Shields claimed, and most of the 1896^97
Midland Chemical Company correspondence assumed, that this cost was
about 1¿ per pound, though Cooper at one point claimed that it might be
as high as 5¿ per pound. In 1910 Herbert Dow estimated that the labor,
packaging, and fuel to convert bromine to potassium bromide would cost
Dow about 3.9¿ per pound. Unfortunately, the extant production records
do not include any information on Dow's conversion costs once it did
purchase Ohio River bromine. Since Dow would have used the same
process as Mallinckrodt and P&W, but with less experience in doing so,
Dow's projections were probably above the costs of the established
distributors. Nevertheless, I round this estimate up and assume a t of 4¿.
III(iii).
A potassium bromide demand function
For simplicity, I estimate a linear demand function for potassium
bromide,
PKBr ˆ a ÿ GQKBr ‡ ui
The demand function was estimated using annual data between 1880 and
1914. Weekly data are available for price, but not output. In any case,
using annual data minimizes the problems created by speculative demand
in response to publicly observed breakdowns in collusion, by averaging
ß Blackwell Publishers Ltd. 1997.
126
margaret c. levenstein
over periods of collusion and price war during a year. Because I do not
have su¤cient data to estimate supply behavior, and so estimate demand
as part of a system of equations, I instead use instrumental variables to
correct for the dependence of QKBr on ui . Annual data on industry output
was taken from the US Geological Survey's Mineral Resources of the
United States. It is undoubtedly measured with error, as there are
discrepancies between measures of output in industry correspondence and
in MRUS. Industry correspondence does discuss the MRUS reports, and
there does not appear to be any systematic bias in the measure. The
potassium bromide price is the average of weekly prices reported in the
Oil, Paint and Drug Reporter. These prices were the prices paid by
pharmacies and pharmaceutical jobbers to the potassium bromide
distributors. I have con¢dence that OPDR prices do re£ect transaction
prices as correspondence often referred to quotations in the OPDR. All of
the discussion of price changes in the industry correspondence is
immediately re£ected in OPDR price quotations. I use two instruments
which are correlated with QKBr but should be independent of ui : the price
of potash, Ppotash , and the percentage of the year the industry was
colluding, Iavg : Ppotash is the annual average of the price announced weekly
in the OPDR. Iavg is the annual average of a weekly indicator series, It ,
which takes the value 1 if the industry is colluding, 0 otherwise.10 The
series is based on weekly reports in the OPDR and correspondence among
industry participants. The estimated demand equation is
PKBr ˆ 60:551 ÿ 0:0388 QKBr
…5:496†
…2:629†
Using the demand equation, I calculate prices, output levels, and pro¢ts
for counterfactual industry behaviors (Table I). Assuming that there were
twelve bromine manufacturers, two distributors, and no potential entry,
the very best that all these ¢rms together could do for themselves was to
set the price of potassium bromide at 38.6¿ per pound, generating about
$125,000 pro¢t, to be distributed among the fourteen ¢rms.11 The average
price of potassium bromide during the pool period (1885^1902), excluding
price wars, was 38.57¿. If, instead of organizing the pool, both levels of
the industry had played Cournot, the price of potassium bromide would
have been 33.5¿; if both had played Bertrand, the price would have been
16.6¿. The pool structure allowed the ¢rms in the industry to increase
10
If one believed that switching between collusive and price war behavior in this industry
was exactly that described in Green and Porter, so that switching was induced by exogenous
shocks to demand, Iavg would be dependent on ut and not an appropriate instrument in this
regression. However, there is no indication in the correspondence that demand shocks were
the source of confusion about whether a ¢rm had cheated.
11
Either of the Bertrand-Joint Maximization boxes in Table I will give this industry
outcome, but the distribution of pro¢ts is very di¡erent in the two cases.
ß Blackwell Publishers Ltd. 1997.
price wars and collusion in bromine pre-world war i
127
Table I
Table of Counterfactual Prices and Quantities
(quantities in thousand pounds)
Demand Curve for Potassium Bromide: PKBr ˆ 60:55 ÿ 0:0388 QKBr
Demand Curve for Bromine, if distributors are Bertrand competitors: PBr ˆ 63:7 ÿ 0:0537 QBr
Demand Curve for Bromine, if distributors are Cournot competitors: PBr ˆ 63:7 ÿ 0:0806 QBr
Demand Curve for Bromine, if distributors are monopolists: PBr ˆ 63:7 ÿ 0:1074 QBr
Upstream (Bromine) Industry Structure (12 ¢rms)
Bertrand
Cournot
Joint Pro¢t
Maximization
(among upstream
bromine ¢rms)
Downstream
(Potassium
Bromide)
Industry
Structure
(2 ¢rms)
Bertrand
PBr ˆ
12.0¿
16.6¿
PKBr ˆ
QBr ˆ 962.9 lbs.
QKBr ˆ 1132.8 lbs.
0
IIBr ˆ
IIKBr ˆ
0
PBr ˆ
16.0¿
PKBr ˆ
20.0¿
QBr ˆ 888.3 lbs.
QKBr ˆ 1045.1 lbs.
IIBr ˆ
35.5
IIKBr ˆ
0
PBr ˆ
37.8¿
PKBr ˆ
38.6¿
QBr ˆ 481.4 lbs.
QKBr ˆ 566.4 lbs.
IIBr ˆ
124.6
IIKBr ˆ
0
Cournot
12.0¿
PBr ˆ
31.2¿
PKBr ˆ
QBr ˆ 641.9 lbs.
QKBr ˆ 755.2 lbs.
0
IIBr ˆ
IIKBr ˆ
110.3
PBr ˆ
16.0¿
PKBr ˆ
33.5¿
QBr ˆ 592.4 lbs.
QKBr ˆ 696.7 lbs.
IIBr ˆ
23.7
IIKBr ˆ
94.1
PBr ˆ
37.8¿
PKBr ˆ
45.9¿
QBr ˆ 320.9 lbs.
QKBr ˆ 377.9 lbs.
IIBr ˆ
82.8
IIKBr ˆ
39.3
12.0¿
PBr ˆ
38.6¿
PKBr ˆ
(among downstream QBr ˆ 481.4 lbs.
bromide distributors) QKBr ˆ 566.4 lbs.
IIBr ˆ
0
124.6
IIKBr ˆ
PBr ˆ
16.0¿
PKBr ˆ
40.3¿
QBr ˆ 444.4 lbs.
QKBr ˆ 522.6 lbs.
IIBr ˆ
17.8
IIKBr ˆ
106.1
PBr ˆ
37.8¿
PKBr ˆ
49.5¿
QBr ˆ 240.7 lbs.
QKBr ˆ
284 lbs.
IIBr ˆ
62.1
IIKBr ˆ
31.3
Joint Pro¢t
Maximization
Note: QBr is always equal to :85 QKBr because of the Leontief technology. It would be impossible for the
potassium bromide distributors to buy less, and not cost minimizing to buy more.
pro¢ts to very nearly the monopoly level.12 The pool's exclusive selling
contracts also foreclosed competition at the distributor level which might
have arisen in the ``Cournot^Cournot'' scenario where the distributors
captured almost 80% of industry pro¢ts.
I can also compare the prices received by the bromine producers during
the Shields pool with those under various counterfactual alternatives. If
the twelve bromine producers had been able to cooperate perfectly and
12
The total deadweight loss induced by collusion, using the average (non-price war) pool
price of 38.6¿, was about $62,219 per year. While for the average consumer this amounted to
virtually nothing (about one twelfth of a penny per capita in 1900), it is about 28% of
predicted average industry revenues.
ß Blackwell Publishers Ltd. 1997.
128
margaret c. levenstein
achieve the joint-pro¢t maximizing output level, they would have received
a price of 37.8¿ for a pound of their liquid bromine. They did not receive
anything close to that under the Shields contracts; the highest price
received was 22¿. That was, however, substantially above the price they
would have been able to achieve if they simply played Cournot (16¿). In
fact, 16¿ was exactly the price they received in the ¢rst year of the Shields
contracts, with prices increasing each year as cooperation was
maintained.
The Midland Chemical Company similarly pro¢ted from the
arrangements it made to cooperate with the pool distributors. From 1897
to 1902, Midland (after 1900, Dow) received 25¿ for each pound of its
potassium bromide, while its average cost per pound was about 8¿. An
explicit counterfactual analysis is more di¤cult in this case, because it is
unclear what industry structure would have obtained as the industry
adjusted to the new entrant and to its lower cost technology. But with a
markup over cost of over 200%, the contracts with Mallinckrodt and
P&W certainly rewarded Midland generously for its agreement not to try
to ¢nd out what that structure might be.
After 1902, Dow's average cost increased, as its new contracts required
that it produce a wider variety of bromide salts than had previously been
the case. Between 1902 and 1905, Dow's bromide cost averaged 1014¿.
Dow's contract speci¢ed that it receive 80% of the sales price (the OPDR
price given the understanding with Mallinckrodt). With an average price
of 28.7¿, the price-cost margin during cooperative periods between 1902
and 1905 averaged 12.7¿, more than a 100% markup over cost.13 Collusion
was even more successful after the agreement between Dow and the Ohio
River was signed in 1910. The average price of potassium bromide was
33.1¿, so that Dow's markup on its cost was over 150%. Some of this pro¢t
was shared with the Ohio River producers. Dow bought Ohio River
bromine at a price tied to the price of potassium bromide, so that the
average price Dow paid the Ohio River producers between 1910 and 1914
was about 17.8¿. Dow's successful entry had eroded their bargaining
power, so that the price of bromine was lower than during the Shields
pool, but it was still above what they would have received as Cournot
competitors.
Finally, while the counterfactual calculations in Table I assume constant
costs over time, it is also possible to compare £uctuations in prices with
£uctuations in costs. Figure 2 displays the £uctuations in the cost of a
pound of potassium bromide during the Shields pool. Fluctuations in cost
are driven by changes in the price of potash and annual increases in the
13
In reality, the price-cost margin was greater than this. This calculation compares the
average cost of producing all bromides with the price received on potassium bromide. The
prices received on other bromides were substantially higher.
ß Blackwell Publishers Ltd. 1997.
price wars and collusion in bromine pre-world war i
129
Figure 2
Potassium Bromide Cost Function Using Ohio River Bromine: MCKBr ˆ bPbromine ‡ pPPotash ‡ t
Note: b ˆ :85, p ˆ :6, t ˆ :04: PBROMINE is the price paid by Mallinckrodt and P&W, as
speci¢ed in pool contracts. PPotash is the annual average of weekly prices listed in the Oil, Paint
and Drug Reporter.
price of liquid bromine, speci¢ed in Shields' contracts. Figure 2 compares
the distributors' cost with the market price of potassium bromide. The
average price during this period was 40.9¿, while the average cost was
24.l¿.14 The distributors' average markup over cost during the Shields pool
was almost seventy percent.15 Fluctuations in the cost of potassium
bromide do not explain changes in its price. The correlation between
changes in price and cost is not even positive (ÿ0.042).
Thus, it is clear that the pool allowed the ¢rms in the industry to
14
The high average price for potassium bromide during the Shields pool may re£ect an
increase in demand, which would have increased the joint pro¢t-maximizing price. But
industry participants much more frequently complained of stagnant demand than mentioned
any increase. Thus it is likely that the increased price also re£ects the e¡ect of double
marginalization on industry price and output levels. Using the prices Shields charged the
distributors for liquid bromine, and the potash prices that prevailed over the same period the
MCKBr was 24.1¿. That increases the distributors' joint pro¢t-maximizing price to 42.3¿.
15
A similar comparison could be made between the distributors' cost using Dow bromides.
Between 1897 and 1902, Mallinckrodt and P&W paid Dow 25¿ per pound, and also incurred
distribution costs. Thus their pro¢t on Dow bromides was somewhat less than on Ohio River
bromine.
ß Blackwell Publishers Ltd. 1997.
130
margaret c. levenstein
increase prices and pro¢ts above any reasonable counterfactual noncollusive behavior. For much of the time, they were able to charge prices
very close to the monopoly level. Despite the heroic assumptions necessary
to compute counterfactual prices we can have con¢dence in them as they
correspond remarkably closely with relevant prices set over the course of
the industry's collusive history. For example, when Dow, having agreed to
divide up world markets with the Deutsche Bromkonvention, wanted to
force the Ohio River producers to contract directly with itself, Dow set the
price of the downstream product so that the upstream ¢rms would have
no incentive to vertically integrate downstream. Dow set PKBr at 19¿ per
pound for almost two years, just below the price …PKBr ˆ 20¿) that would
have obtained if the Ohio Rivers played Cournot and converted and
distributed their output themselves, as some of them attempted to do (the
Cournot-Bertrand combination in Table 1).
iv.
comparing models of price wars
Six price wars disrupted cooperation during the period of this study
(Figure 1). Each of these price wars was described as such in the Oil, Paint
and Drug Reporter which regularly and explicitly discussed pool activity.
Three price wars preceded Dow's entry into the industry (Pool Wars I, II,
and III), and three followed (Dow Wars I, II, and III).
Friedman [1971] demonstrated that o¡-the-equilibrium-path threats of
permanent Cournot reversionöa price waröcould induce collusion in a
su¤ciently patient industry. But in Friedman's equilibrium no price war
was ever observed. In contrast, Green and Porter [1984] and APS argue that
the successful use of punishment threats in an industry where cheating is
unobservable will occasionally require their implementation. When ¢rms
cannot directly observe deviations from prescribed strategies (``cheating''),
they rely on observations of a public variable (e.g. price) whose distribution
depends jointly on the actions of each ¢rm (e.g. each ¢rm's output) and a
random disturbance. When the public variable falls into a certain range
(e.g. prices fall below some p ) ¢rms revert to punishment behavior (e.g.
higher output levels). Since ¢rms cannot distinguish between cheating and
low realizations caused by random £uctuations, they must punish to keep
incentives intact.
Many aspects of the collusion in bromine resonate with the intuition of
this class of games. First, the repeated interaction of ¢rms was crucial to
the success of cooperation. The clearest evidence of this is the
instantaneous breakdown of cooperation between the Americans and
Germans upon the outbreak of war in 1914. The war prevented the
Germans from exporting their goods. When the threat to punish cheating
in the present with low prices in the future became empty, collusion
between the Americans and Germans was no longer sustainable.
ß Blackwell Publishers Ltd. 1997.
price wars and collusion in bromine pre-world war i
131
Second, one agreement between the Dow Chemical Company and
the Deutsche Bromkonvention explicitly included ``equilibrium punishments.'' The contract prohibited the parties from exporting into each
others' exclusive territory. It provided speci¢c penalties, far short of
permanent Cournot reversion, if such exports (or their observational
equivalent) did occur. The speci¢ed punishment was a simple tit-fortat rule (Axelrod [1984]). It reads,
Should, contrary to agreement, American bromin or its salts arrive
in Europe during the continuance of this agreement the Germans are
at liberty to send the corresponding quantity of bromin, or its salts,
for sale to the North American market . . . Should European bromin,
or its salts, be shipped to North America the Americans will be at
liberty to send to Europe the same quantity of bromin or its salts to
the same equivalent as was imported into North America.
While there were instances of bromine products exported into exclusive
territory (usually by a third party), this punishment was never
implemented. Instead, the parties would agree on a dollar amount to be
paid by the presumed violator to the aggrieved party. The penalty was
usually equivalent to the presumed pro¢t on the transaction, though in
one case of an egregious violation that Dow acknowledged having
committed, an additional penalty was imposed.
Third, bromine industry participants could only imperfectly monitor
one another. Three kinds of cheating were possible. Bromine producers
could sell outside the pool. The distributors could sell bromides below the
agreed upon price. Or they could sell bromides in markets that were
someone else's exclusive territory. Even if no regular participant in the
bromine industry engaged in any of these activities, the storability of
bromine meant that third parties could sell goods purchased during an
earlier price war. The possibility of such speculative activity increased the
di¤culty of monitoring compliance. The bromine pool relied on three
methods for detecting possible violations: 1) secret, direct observation of
plant activity (spying), 2) monitoring £uctuations in customer demand,
and 3) monitoring government reports on imports and exports. None of
these methods provided perfect information.
Finally, price wars threats were used to induce cooperation with the
cartel. Correspondence among industry participants is replete with dire
warnings of imminent price wars if the reader did not cooperate. Internal
Midland Chemical Company documents are explicit that such threats
induced the company's cooperation. Shortly after it began producing
bromides, Shields threatened to resume the price war of the previous year
if Midland did not agree to sell its entire output to him, on the same terms
as the Ohio River producers had. Midland agreed to restrict its output
ß Blackwell Publishers Ltd. 1997.
132
margaret c. levenstein
and sell only to Mallinckrodt and P&W, despite Herbert Dow's stated
preference to sell independently of the pool.
Similar concerns explain Midland's continued cooperation with the
pool. Because its technology had increasing returns, Midland was anxious
to increase its output.16 Between 1894 and 1896, Midland's output quota
increased from 100,000 to 150,000 pounds, far below its capacity of over
350,000 pounds per year. When negotiations began in 1896 to renew the
contracts with Midland and the Ohio River producers, Midland
considered breaking from the pool. The contract Midland eventually
signed limited its annual output to 200,000 pounds. The acceptance of a
contract with which it was clearly unsatis¢ed was the result of its belief
that P&W and Mallinckrodt had accumulated 800,000 pounds of bromine,
which they had threatened to dump on the market if Midland did not
agree to cooperate. Midland was afraid that the pool would bankrupt the
young ¢rm, and so chose instead to cooperate.
Thus in many respects the intuition of Green and Porter and APS is
borne out in the bromine case. There are di¡erences between these two
models; with the data available, it is possible, in a limited way, to compare
their applicability to the bromine case. Green and Porter assumes that
¢rms will revert to one-shot Cournot behavior during price wars, and
predicts that price wars will either be permanent or last a ¢xed number of
periods. APS, on the other hand, predicts that ¢rms will switch back and
forth between two states: maximal collusion and maximal punishment
(potentially more severe than Cournot reversion).17 The simplicity of the
APS optimal strategy, in which ¢rms play one of two actions, depending
only on the prior period realization of the public variable, is appealing
both for its analytical elegance and the ease with which such a simple
strategy could actually be implemented.
These qualitative characteristics of APS price wars (described in Table
II) seem at variance with those observed in the bromine industry. The APS
optimal punishment prescribes the same behavior across price wars and
the same behavior in every period of the price war. There is, however,
substantial variation in the characteristics of the bromine price wars. The
initial fall in price ranges from 1¿ to 15¿, or 3% to 50% (Table II). The
16
An obvious question is why P&W and Mallinckrodt did not purchase their entire
requirements at a lower price from Midland. First, the Ohio River ¢rms would not
immediately exit. Second, P&W and Mallinckrodt did not want to accept the risk of having a
sole supplier of their bromine requirements.
17
This implies that, in the APS case, the state of the industry follows a ¢rst-order Markov
process. I conducted several tests of whether the state of collusion in the bromine industry
followed a Markov process. While these tests could not reject the hypothesis of a ¢rst-order
Markov, they were largely inconclusive. More importantly they could not distinguish between
a ¢rst-order Markov process and a series generated by a much more complex set of strategies
in which the probability of a price war was lower immediately following the implementation
of a collusive agreement. See the Journal's editorial Web site for details of these test results.
ß Blackwell Publishers Ltd. 1997.
price wars and collusion in bromine pre-world war i
133
Table II
Characteristics of Price Wars Compared
Price War
1886^1887
1888
1891^1892
1902
1903
1905^1908
Price war
Name
Pool I
Pool II
Pool III
Dow I
Dow II
Dow III
Initial Drop
(¿)
2¿
1.5¿
1¿
4¿
*
15¿
Initial Drop
(%)
6.0%
4.4%
3.0%
9.1%
*
50%
Total Drop
(¿)
2¿
1.5¿
13¿
19¿
*
17¿
Total Drop
(%)
6.0%
4.4%
39.4%
43.2%
*
56.7%
Duration
(weeks)
22
12
82
6
21
201
Weeks to
trough
1
1
73
6
*
149
Low price
32¿
32.5¿
20¿
25¿
*
13¿
* There was an 18¿ price spread in the quotations for potassium bromide during this Dow War II.
Virtually no quantities were exchanged. This was the only time during the period studied when there was
any spread in published price quotations.
Note: All price data are from the Oil, Paint and Drug Reporter. Price wars periods are determined
independently of prices, from reports in the OPDR and in internal industry documents.
duration of price wars varies from 6 to 201 weeks, the total drop in price
from nothing to almost 60%. This variety may re£ect stochastic
£uctuations as well as structural change in the industry. But some price
wars also display change in behavior within a single price war, with prices
spiraling down over the course of the war. In these cases, there was not a
switch to a single, commonly understood, worst punishment.
The bromine industry did not use punishments more severe than
Cournot reversion. In four of six wars, the price decline was not large
enough to reduce the expected present discounted value of the ¢rm's future
pro¢t stream below the Cournot reversion level. Only the low prices of
Pool III and Dow III wars reached the one-shot Cournot level.18
(Compare Tables I and II). Prices never fell below marginal costs.19
There are several reasons why the bromine price wars were generally less
18
The low price during Dow War III is 13¿, less than the 20.5¿^23.6¿ range for the Cournot
price reported in Table I. But the cost function presented re£ects the costs of the Ohio River
producers. Dow's costs were much lower so the estimates in Table I no longer apply in 1905,
when Dow produced more than half the US output.
19
This is true as well for Dow War III, when prices are compared to Dow's costs. The
low price reached was 13¿, but Dow's marginal cost was less than 10¿.
ß Blackwell Publishers Ltd. 1997.
134
margaret c. levenstein
severe than those contemplated in the theoretical literature. First, pool
members discounted threats which they believed too severe to be credible.
Dow dismissed claims that other ¢rms would lower price below their own
costs. Dow demanded veri¢cation of the inventories Mallinckrodt and
P&W claimed to have ready to dump on the market. Threats designed to
support collusion were useful only when credible. When imperfect
information led to the implementation of such threats, the ensuing price
wars were relatively mild.
Second, the possibility of storing bromine for future use, or for re-sale,
increased the costs of price wars. At the end of any price war, the demand
faced by pool members remained low until speculative output had been
consumed and the market price had reached the pool price. The increased
costs of engaging in price wars with storable goods decreased the severity
of the optimal price war punishment. This issue was not unique to the
bromine industry, however, or even to manufactured goods industries. For
example, the JEC's customers, understanding that there was a possibility
of a price war among railroad lines, would store their grain in the west in
anticipation of a price cut (Ulen [1979], pp. 165^166). Thus the possibility
of arbitrage, by customers or third parties, makes it likely that the optimal
punishment will be less severe than that contemplated in APS.
Finally, there seems implicit in each price war a reputational cost, such
that if other ¢rms believed that some participant had cheated, collusion in
the future would be harder to sustain.20 This implicit cost may have made
it possible for relatively mild punishments to sustain collusion.
Thus, the qualitative characteristics of the bromine price wars seem
to di¡er in important ways from those implied by APS strategies. In
particular, equilibrium punishments more severe than Cournot were not
used in this industry, for reasons that are common to many other
industries. Even in the most e¡ective bromine cartel, established by
Dow and the Deutsche Bromkonvention, the written agreement did not
pretend to call forth punishments as severe as contemplated in the
theoretical literature. In practice, even those punishments were often
eschewed as the two parties set up mechanisms to provide themselves
with more information about, and therefore, to deter, cheating. They
also collected information that allowed them to identify probabilistically
the likely perpetrator of any cheating. They then introduced ¢rm-speci¢c
punishments, transferring pro¢ts from one ¢rm to another, rather than
using punishments that lowered pro¢ts to the industry as a whole.
Although ``simple'' strategies have analytical elegance, ¢rms in the
bromine industry were happy to avail themselves of more complex
strategies if necessary to collude successfully.
20
While their model is di¡erent from those discussed here, the intuition is similar to that
in Kreps, Milgrom, Roberts, and Wilson [1982].
ß Blackwell Publishers Ltd. 1997.
price wars and collusion in bromine pre-world war i
135
There were two price warsöPool III and Dow IIIöin which prices
did fall below Cournot levels (Tables I and II). But this severity seems to
have arisen from a di¡erence in the underlying cause of the price wars,
rather than in the severity of the optimal punishment. These two price
wars were more like Stiglerian collusive breakdowns than ``imperfect
monitoring'' price warsöwhich arise not from real cheating, but from
indistinguishable stochastic shocks. While imperfect information
problems exacerbated and lengthened these two price wars, they both
began with very public announcements that participants were violating
the existing agreement. Dow War III began with a public announcement
that the Germans were willing to sell below the pool price, in response to
Dow's very real exports to Europe in violation of their understanding
with the Germans. Similarly, Pool War III began with a public
renunciation of the existing agreement by the Ohio River bromine
producers. In both cases, industry participants understood that these
announcements were part of a strategy to renegotiate the distribution of
rents. In these cases, the limits on the severity of price wars did not
obtain as prices fell nearly all the way to marginal costs. These price wars
were part of the negotiation of a new collusive arrangement, not a
punishment phase that would help to re-establish an existing one. At least
in the case of Dow War III, the collusive agreement broke down
following a change in the relative costs of ¢rms in the industry.
Competition during the price war revealed information about ¢rm costs
and staying power, and thus provided the ground rules for the
negotiation of a new collusive agreement.
conclusion
From 1885 to 1914, participants in the bromine industry colluded,
increasing prices and pro¢ts. Some of the mechanisms used to sustain this
collusion resemble those described in the repeated game literature. As
suggested by Friedman [1971], threats of price wars helped sustain
collusion. Imperfect monitoring problems contributed to at least some of
the six price wars observed. While there are similarities to the punishment
strategies proposed by Abreu, Pearce and Stacchetti [1986], the bromine
price wars were not as severe as they contemplated. Arbitrage increased
the cost and limited the severity of imperfect monitoring price wars.
Threats of severe price wars were not considered credible. The most severe
price wars seem to have resulted not from monitoring problems, but from
the more basic disagreements which inherently plague attempts to collude.
The bromine case provides evidence in support of Green and Porter's
[1984] claim that the occurrence of price wars is not inconsistent with
successful collusion. But if game-theoretic metaphors of equilibrium price
ß Blackwell Publishers Ltd. 1997.
136
margaret c. levenstein
wars are to have meaning, they must be disguishable from price wars
which re£ect a breakdown in collusion. The bromine industry provides
examples of both.
MARGARET C. LEVENSTEIN
Department of Economics,
University of Michigan,
Ann Arbor, MI 48109-1220,
USA
email: [email protected]
ACCEPTED OCTOBER 1996
references
Abreu, D., Pearce, D. and Stacchetti, E., 1986, `Optimal Cartel Equilibria with
Imperfect Monitoring,' Journal of Economic Theory, 39, pp. 251^269.
Axelrod, R., 1984, The Evolution of Cooperation, Basic Books, New York.
Burns, A. R., 1936, The Decline of Competition: A Study of the Evolution of
American Industry (McGraw-Hill, New York).
Davis, L. E., 1965, `The Investment Market, 1870^1914: The Evolution of a
National Market,' Journal of Economic History, 25, pp. 355^399.
Friedman, J. W., 1971, `A Non-cooperative Equilibrium for Supergames,' Review
of Economic Studies, 38, pp. 1^12.
Green, E. J. and Porter, R. H., 1984, `Non-cooperative Collusion Under Imperfect
Price Information,' Econometrica, 52, pp. 87^100.
Haynes, W., 1954, American Chemical Industry: Background and Beginnings v I.
(D. Van Nostrand, New York).
Kreps, D., Milgorm, P., Roberts, J. and Wilson, R., 1982, `Rational Cooperation
in the Finitely Repeated Prisoners' Dilemma,' Journal of Economic Theory, 27,
pp. 245^252, 486^502.
Levenstein, M. C., 1993, `Price Wars and the Stability of Collusion: A Study of
the Pre-World War I Bromine Industry' DAE Working Paper 50, National
Bureau of Economic Research, Cambridge, Mass.
Levenstein, M. C., 1995, `Mass Production Conquers the Pool: Firm Organization
and the Nature of Competition in the Nineteenth Century' Journal of Economic
History, 55, pp. 575^611.
Levenstein, M. C., 1997, Accounting For Growth: Competition, Information
Systems, and the Creation of the Large Corporation, Stanford University Press,
Stanford, CA.
Porter, R. H., 1983, `A Study of Cartel Stability: The Joint Executive Committee,
1880^1886,' Bell Journal of Economics, 14, pp. 301^314.
Slade, M. E., 1991, `Market Structure, Marketing Method, and Price Instability,'
Quarterly Journal of Economics, 106, pp. 1309^1340.
Stigler, G. R., 1964, `A Theory of Oligopoly,' Journal of Political Economy, 72,
pp. 44^61.
Stocking, G. W. and Watkins, M. W., 1951, Monopoly and Free Enterprise,
Twentieth Century Fund, New York.
Ulen, T. S., 1979, Cartels and Regulation: Late Nineteenth Century Railroad
Collusion and the Creation of the Interstate Commerce Commission. Ph.D.
dissertation: Stanford University.
ß Blackwell Publishers Ltd. 1997.
price wars and collusion in bromine pre-world war i
137
US Census, 1880, `Chemical Products and Salt,' Report on the Manufactures of
the United States at the Tenth Census, GPO, Washington, DC.
US Department of Commerce, 1975, Historical Statistics of the United States:
Colonial Times to 1970, Part 1 (GPO, Washington, DC).
US Geological Survey, 1883^1915, `Bromine,' Mineral Resources of the United
States, GPO, Washington, DC.
Bibliographical Note
The Herbert H. Dow Papers are housed in the Post Street Archives, Midland,
Michigan.
The Oil, Paint and Drug Reporter was ¢rst published in 1871. It has since been
superseded by the Chemical Marketing Reporter.
ß Blackwell Publishers Ltd. 1997.